chunked/nltk_chunking/model_doc_perceiver/chunk_6.txt

This means, for example,
Perceiver IO can do BERT-style masked language modeling directly using bytes instead of tokenized inputs.